I did something like this once many years ago for one of my wife's students, a precocious high-schooler who needed to be challenged beyond the elementary Fortran she was teaching at the vocational school. So I outlined the steps it would take to translate a toy language into a toy computer's instruction set, and reviewed it with the young man one afternoon at my office - an abbreviated compiler theory class with an eager audience of one. Tom took the big flowchart that we made home, studying and meticulously documenting it. When he brought me the documented design a couple of days later I asked him to implement it.

Tom spent awhile doing this in Waterloo Fortran, and when he presented me with the working compiler I said "Good job. Now execute the code." He blinked his surprise - until then I don't think he'd considered that the toy machine could have been interpreted. A revelation! He went back home and wrote the VM for the toy machine without any assistance; it worked well and I was as proud of him as I could be.

A couple of years later I hired Tom and a couple of years after that he lapped me, going on to better things. 30 years later I still keep a copy of his compiler and interpreter, reminding me from time to time of the bright kid who grew up to become a fine programmer and family man.