I built a toy CPU in software once as an exercise. I started with "class Transistor" (wrapping an AND op) and "class Wire" (wrapping a boolean), and wired them together incrementally to make gates, flipflops, registers, etc.

I eventually got a fully-functioning 32-bit cpu with instruction pipelining, two levels of cache, DMA input/output, an asynchronous bus, a custom assembly language with an assembler written in python, and got the Game of Life running on it.

It ran about 2kHz with 8kb of memory or so.