Looking at this, I'm not seeing a lot that can generate elaborate stochastic structures. As a synth, it's an excellent build, but I'm thinking it'll be rather limited for complex generative work, particularly something self-regulating. Have you perhaps considered a second cab to tandem with this one that can handle that level of complexity? Yes, less CAN = more...but for what you're envisioning, it seems a little TOO 'less'.
Consider something that adds some complex stochastic CV pattern generation, like a Catalyst Audio Time's Arrow, then tandem that with some methods for altering clocking behavior that includes both a stochastic element (such as a Euclidian sequencer) and logic functions that complicate potential trigger/gate patterns, which would work nicely with the Pamela's. Lastly, consider some sort of modulator 'cascade', in which you have some low-speed modulators that can continuously vary some downstream LFO or envelope sources, or a function generator such as a Maths, and maybe tandem this with a quantizer + shift register to 'grab' voltage curves for quasi-random arpeggiations. So that way, you'd have a top cab for 'voice' and the lower for 'control'. With something like that, you could also add some sound processing and also subject that to generative control.