I went from Class D to Luxman A/AB - And most of what you think is wrong

Eric, very interesting.  I had similar situation switching from Rowland 102, an early Icepower (200ASC) class D amp, to Benchmark AHB2 class A/B.  Rowland was a little bit darkish sounding while Benchmark is vivid and airy with better treble extension.  What surprised me the most was the bass extension - much better with Benchmark.  I cannot tell if class D sounded worse when cold, since I kept it on all the time,  but I remember that it took initially long burn in time (about 400hrs) to make it sound the best.  Benchmark is better sounding overall, but it is for 2x more money.  It would be interesting to compare Benchmark to class D amp with the same price tag.

One more thing (as Columbo said): SMPS are not cheap. Cost of high quality SMPS has to include high cost of development. Rowland’s supplies deliver thousands of regulated watts operating at 1MHz (and components doing this are not cheap either). I couldn’t believe it is even possible, since most of switchers operate at 1/10 of this frequency. I can give you a simple reason (other than complexity) why many manufacturers decide to go with linear supplies. It is because there is still a lot of people thinking like you.  There was similar prejudice against all class D amps not so long ago.  Many believed they are good only for subwoofers.  I would advise the same as Eric did - don't look too much into technicals like class of operation or type of supplies, but listen instead.

willgolf, My class D amp was designed to be on all the time (no power switch).  I kept it on with surge protector, but unplugged it when on vacation or during thunderstorms.  You could call Mola Mola to be sure.

Class D can be wonderful but in my experienced opinion, if it has a SMPS just forget it! Even my Mac Mini music server SMPS killed the sound, it now has a linear supply.

All highly rated Rowland class D amps have SMPS. When done properly SMPS is much quieter than linear power supply, not to mention line and load regulated while linear supply is not.
Rowland uses SMPS even in preamps, where efficiency is not important, just because they can be so quiet. Another example is Benchmark. Their products, known to be extremely quiet, contain SMPS. They were able to lower noise by 10dB in their DACs just by switching from linear supply to SMPS. AHB2, an extremely quiet power amp uses SMPS.

"Linear Power Supply", in fact, is a very primitive unregulated switcher operating at 120Hz and switching at max voltage, charging output capacitors with narrow current spikes of very high amplitude. It produces 120Hz ripple as well as very narrow switching spikes, both very difficult to filter out. Also, many believe that transformer has to be heavy to deliver a lot of current. It is all matter of frequency. Small 1" ferrite transformer operating at 100kHz can deliver as much power as huge transformer operating at 60Hz.

Yes, SMPS can be very bad, especially when in cheap/crude computer power supplies but can also be wonderful when done right. It is funny that anybody can think, that SMPS is not good enough for class D, while class D is modulated SMPS.

Switching amp (class D) and SMPS  IS the same thing.  In fact class D was discovered when engineers tried to show how responsive SMPS was by modulating it with music.  I would say that it should be more difficult to build good class D amp than good SMPS since class D amp has to maintain linearity while SMPS only keeps voltage steady.

As for the noise in amplifier - it can be easily measured.  I currently listen to music coming from my Benchmark AHB2  power amp driven by Benchmark DAC3, both powered by SMPS, and it sounds wonderful. Both received the highest Stereophile ranking in "2019 Recommended Components".

Linear power supply has to have a lot of output capacitance to keep voltage steady under the load and to filter out 120Hz ripple.   These huge, inductive capacitors are in series with the speaker (circuit closes thru them) compromising sound.  

Here is interesting FAQ: 
https://benchmarkmedia.com/blogs/application_notes/152143111-audio-myth-switching-power-supplies-are-noisy

It mentions AHB2 (powered by SMPS), perhaps the quietest power amp ever, with 132dB S/N.

I don't question your bad experience with SMPS understanding where it might be coming from, but keep your options open.  Jeff Rowland wouldn't compromise his business by completely abandoning linear supplies and making everything with SMPS if there is something inherently wrong with it.

Calling it a digital amplifier is a lot closer to the truth than calling an analog amp.

Amplification process is still analog, hence "Analog Amplifier".  Connecting amplifier to DAC doesn't make amplification "digital".

When class D amp is switching between two discrete levels not the voltage but duty cycle is a quantity of interest and it is linear (no discrete steps).  It becomes converted back to corresponding voltage thru averaging (filtering).  Voltage gain of such amp is a function of switched voltage amplitude.  

Also it’s not that the higher switching speed alone make for the good sound it gets, it’s also that the "low order output switching noise filter" that all Class-D’s have to use, can now also be taken up 2.5 x higher, this then has no effects now down into the audible range, like the filters today do, which can cause phase shifts down to 5khz, and why many complain about something not right in the upper mids/highs.

That's true, but it might open another bag of worms.  In current designs filter leaves about 1% of switching noise on the speaker cable.  It is only small percentage but of very high value, reaching 100V - making it approx. 1V peak switching noise on the cable.  It can be easily seen with the scope.  The only reason why it does not radiate is the fact that 500kHz wavelength is 600m.  Antena becomes very ineffective below 1/10 of the wavelength - in this case 60m.  There will be still some, very small electromagnetic radiation but it should not be a problem.  Increasing switching frequency, let say 10x will make this antena "dangerous" at 6m, producing some radiation even with regular speaker cables.  It is possible to filter it better, but it would lower bandwidth resulting in phase shift in audio band, that we want to avoid.  Perhaps compromise is somewhere in-between, increasing switching frequency only 2-3x while still enjoying improved linearity thru reduction of dead time.

You’ll just have to look at what Technics did on the SE-R1, no complaints about it becoming an RF transmitter using normal non shielded speaker cable, just praise for it's good sound.

Perhaps because they raised frequency response only to 90kHz, implying about 50% increase in switching frequency (my old class D amp had 60kHz bandwidth).  They could go higher with GaN Mosfets but stopped at only 90kHz possibly because of RF radiation.  Rowland's integrated Daemon has 70kHz bandwidth at 10x more output power (1500WPC @ 8ohm).  AFAIK it uses plain non-GaN Mosfets.

Yes 90kHz is a little better than 60 or 70kHz, but not by much - leaving noticeable phase shift in audio band. I’m not even sure that this phase lag is a problem, since speaker, being inductive at higher frequencies, shifts phase forward. Lamm Industries Signature Monoblock ML3 has only 30-40kHz bandwidth (that should introduce substantial phase shifts in audio band), but was highly praised by Stereophile: "It was quiet, fast, detailed, dynamic without reservation, transparent, airy, and extended on top".
This example suggests that it is far better to form opinion after listening to particular amp than making blank statements about whole class D.

@georgehifi  
Who said production boards will be soldered by hand?  Many small companies, including one I worked for, make small quantity of very sophisticated high density boards using Assembly Houses.  We used to make them with semi-automatic stencil printer, manual pick and place and 4 zone Heller reflow oven, but using Assembly house  ended up being much cheaper and better, quality wise.  Nobody sane would hand solder production quantities - too much time consuming equals too expensive.  We made batch of about 5 boards for prototyping and then about 100-200 boards in production quantity.  Using good Assembly House opens whole new world for designer (removes constrains).  Now he can design for higher density or even for BGA chips.  

@atmasphere  Ralph, perhaps I can help a little with SMT question (although you may already know all this) .  Suggestion to use wave soldering for your production is nonsense.  It was replaced by reflow soldering more than decade ago.  The only all SMD done with wave soldering today are low density high production quantity boards  (wave soldering is a little faster than reflow).  Even mixed boards are today soldered with reflow ovens + selective soldering (mini solder fountain).  Limitation of wave soldering is at about 20 mil pitch, but board has to be designed for that (orientation of components) and everything else has to be perfect.  It is perhaps good for 25mil pitch only (compromised adhesion of thin solder dam between pins causes shorts).  Also, board has to be designed for wave soldering in mind since recommended component land areas are different - wider (take more space).  I would design everything for reflow.  Hand soldering prototype with 50mil pitch components is easy.  I can do 25mil pitch large chips with 3D magnifier (Mantis) and micro soldering iron, but it is time consuming.  I've seen technicians who can solder 20mils.  For discrete components 0402 is the limit (already hard to rework).  0603 is practical and easy.  Best bet for small operation soldering would be manual stencil and small 4 zone reflow oven (bigger number of zones is mostly for production speed - faster conveyer belt).  Vapor Phase (drawer style) soldering might be also a good option (more accurate).  Better yet leave it to good Assembly House.  They make prototype quantities and small batches.  They often can buy parts at lower cost.  I went thru about 10 assembly houses before I found one that is wonderful.  Location doesn't really matter since you send them zip file and they ship back boards (leave everything to them!).   Novatronix in Chicago area is the best I know.  Sure, it will cost you more in prototype quantity, not to mention set-up charges, but debugging and testing poorly made board can cost you a lot of time (and frustration, and wrong decisions).  I went thru all this.
For your size of operation Altium software is most likely optimal.

Sorry, for all these details, but it goes toward possibility of making own design class D SMT boards vs using standard off the shelf modules.  The answer is - yes it can be done, many ways.