Is it one cc per channel, i think a 16 cc matrix per arp channel would be good. Keep 4 arps with note data and add a cc matrix to sequence for 16 ccs per arp.
Then lfo, velocity and pitch are nice, but LFOs/envelopes need a variety of shapes and control parameters to create interesting non basic shapes. Temporal Warp, amplitude, reverse, invert, duration, loop rate synced and unsynced, phase, fold, Sustain point (reaches a point in shape and holds until the next trigger), loop loops within when sustain point reached if enabled.
Encoders would be useful to draw our own shapes with the ability to stretch the shape over a time division, then have a looping, single one shot, retriggered or unlinked state.
Attenuation, offset and inversion per cc envelope/lfo shape is needed to keep the shape but still be able to attenuate/increase/invert/ positive or negative shift its values.
Ability to assign these ccs LFO shapes internally to midicake arp parameters or externally to devices.
Copy paste as I imagine if you do the above there are a lot of parameters to tweak that may be repetitive.