I have a couple of code projects in C++/Python in which LaTeX-format descriptions and labels are used to generate PDF documentation or graphs made using LaTeX+pstricks. However,
As you're considering using TeX itself for doing the rendering, I suspect that performance is not an issue. In this case you've got a couple of options: dvi2txt to fetch your text from a single dvi file (be prepared to generate one for each label) or even rendering dvi into raster images, if it's ok for you - that's how hevea or latex2html treats formulas.