Jez:
I think that to make an 8052 with subset instruction is a cool idea for the power saving and area saving. But I think for the near term you should consider the following points.

1) Any good C compiler for an 8051 family part is going to be written to take advantage of any trick the instruction set offers in order to try to produce efficient code. So getting a mainstream compiler that has 8 -> 10+ years of development and optimization is just not going to fit against your desires.

2) I suspect that almost any of the C compiler vendors would be interested in re-writing the code generator part of their compiler in order to get you what you want BUT I suspect this is an expensive proposition both in terms of man-months and money spent.

3) I would propose that it could be best to code programs for this specialized processor in assembly language. This way you have full control over what assembly instructions are used. And also since most assemblers are table driven from what I know, it should be cheap to get an assembler vendor to produce a stripped down assembler with the various opcodes stripped out.

4) Have you considered the possibility that you could gain your power savings and chip area reductions by simply stripping out the SFRs and peripherals that are not needed??

5) Have you looked at other processor cores wherein you can get to the desired power versus area density on the chip without having to strip down the code architecture of the cpu section. I used to work for a pacemaker company that some years ago went through the exercise of finding a processor core to put on a pacemaker custom LSI. Pacemakers have to last for 10 -> 12 years and a new battery is not an easy proposition. So their situation was somewhat similar to yours. As I recall they went with a 6502 core as the most power & area efficient.

6) Alternate cores are interesting because you get the chance to keep full CPU definition which leads to the availability of off-the-shelf stable and well tried tool sets.

7) Simpler architecture cores are going to be those processor types that dont have the concept of registers like the 8051's R0 -> R7. Typically these will have an A register, maybe a B register, some index or X register, a PC and an SP, plus an instruction set that treats low 256 memory with simple byte addressing in the instruction set. Cores that I know fit that description are 6800, 6805, 6808, 6303, 6502, maybe 68HC11, and maybe 1802. Notice that surprisingly none of these are Intel derived architectures. Intel's processors have always been more complex as regard to logic structures.

-------
Dare I ask why you are looking at starting with 8051 and stripping as opposed to picking another simpler processor and enhancing as needed to get performance you need within your other constraits budget?

Michael Karas