I would not use FFT data to do such a simple job. Instead, consider running the input signal through 3 simple time domain IIR filters that split the information into low, mid and high signals. This results in 3 time domain signals that are easy to plot. The advantage of this approach is that you do not have to worry about the inverse FFT process that would give you your time domain waveforms from each band. It should also be less processor intensive as you are doing everything in the time domain.
There is some simple code for a 2nd order Butterworth bandpass filter contained in one of my github repos at about line 120 of this file that you are welcome to use as a start point.