General algorithms are generally a bad idea on FPGAs.
They work well on CPUs because storage space is cheap; you can have all the different versions/options and only run the ones you want. On an FPGA, however, you use hardware for all of the versions/options even though you're only actually running one of them - which is very inefficient. Ideally you should also optimize the datatypes and operations used to be FPGA-friendly (eg no floating-point).
I suggest implementing ANFIS in MATLAB first (it's probably got a toolbox for that) and determining exactly what you need - what data types, how many inputs/outputs, what transfer functions. Once you've done that you'll have a much easier problem that you can probably implement in HLS yourself.