Veeerry nice work! That said: if meant to be anything more than an intellectual exercise, imho it's better to target an existing system/architecture. There's a good # of existing systems out there that:a) Can be built from discrete parts (okay, CPU & ROM/RAM excluded - usually). And
b) Have an existing software library. Often a huge one.
b) Is the important bit here. It gives you a full suite of editors, assemblers, compilers, debuggers, productivity software, games, etc etc from day 1. Which bypasses the chicken-and-egg problem of "do something useful with it".
Modern IC's are not black boxes by definition. It's just the scale of today's VLSI that makes inspection by end users impossible.
Even eg. a lowly Cortex-M0 could be considered a complex beast in this context. But buy eg. 100....1000 (8 bit) microcontrollers, take a representative sampling of those (say, a few dozen specimens), decap, put under microscope & compare with architecture documentation. When determined "ok", use the rest of that batch to build stuff. Tedious? Yes! But (for a sufficiently motivated individual or organisation): doable.
Same for small-sized ROM/RAM & peripheral IC's.
IC vs. discrete logic is not the (essential) issue here. Scale/complexity of modern IC's is. Take a # of steps down the order-of-complexity-magnitude scale, and go from there.