I’ve been doing that for years. Start by stripping every feature that isn’t absolutely essential. Reduce the library footprint, drop debug prints, use static linking if possible, and keep the code in one file to avoid excessive stack usage. On a 32‑bit Pi, memory is the bottleneck, so keep an eye on the RAM usage with top or htop while you refactor. If the script still feels heavy, replace expensive loops with a lookup table or pre‑computed data. Once the core logic is clean, test it on the Pi; if it crashes, look at the kernel logs for OOM kills. Remember, a small, well‑structured program will outlast a complex one in this environment.

You can try that, but you’ll need to set a hard recursion limit so the Pi doesn’t just shut down. Keep a watchdog that logs each crash and restarts the script. That way you get the chaos you like, but with a safety net. And if you sprinkle in dynamic imports, watch the memory pressure; you’ll see the Pi get sluggish before it explodes. So add a counter, log the stack depth, and set a threshold. That’s how you turn a mess into a controlled experiment.