In principle, a convolver is most similar to a delay in terms of buffer management, except that it uses them sideways, in a way. The buffer may just have to be the length of the impulse divided by the duty cycle length. That means the impulse length divided by duty cycle length determines the maximum size of the buffer. Each duty cycle samples its fraction of the impulse sampled, multiplied by the buffered sample from the index into the impulse sample, determined by multiplying its position in the buffer times the duty cycle length. Then the buffer gets transported cycle by cycle until the end. It's a tiny bit of a brain teaser to me, but there are worse out there.
There's the "easy" way, which would simply offer the loading of an impulse sample and then some parameters such as length percentage as minimum, wet/dry (maybe) and what ever else could be interesting, which might not be much.
The really amazing way would be two audio inputs, where you could essentially feed anything into the 2nd (impulse input), while you'd determine the length in the convolver, for example.
This opens up a whole new world of possibilities, naturally, and gives MUX yet an extra edge in terms of usability. Seems like there's some niche war going on in the convolution plugin world and it would be quite a statement to declare independence!
