It certainly would speed things up (the current TI-BASIC interpreter is more or less just a expression evaluator that has had various things haphazardly bolted on to it over the years to turn it into a usable language- lots of things are done that are nothing short of silly), and yes it would be
possible to make such a compiler, especially if you make it a standalone tool instead of something run on-calculator.
Is it worth it? Probably not, for as much work as it would be.
You may be interested in PT_'s
IndiumCE project, which aims to write a faster
interpreter that pre-parses the program (similar to how Python and many modern interpreters work), among other significant optimizations. It currently exists in more of a prototype stage, if I'm not mistaken, so don't expect it to work on most programs.