The interpreter I'm building right now will essentially be a JIT, but it will JIT to an internal IR instead of machine code. My plan is to eventually have it operate on compact and optimized stack frames as native code would.