There's no discussion section so I'll write in an issue. Just found this repository so I didn't check the code yet. But looks good.
Some questions and comments:
Any plan to use Vector128 and Vector256, SSE and AVX respectively, to speed up the library?
How do you plan to support trigonometrics functions?
Have you looked at double double 128-bits floats? It uses two IEEE754 double to create a 128 bits float. It's not IEEE754 compliant but it does offer more precision than a double.