This is not something we recommend tracking with Prometheus.
The problem is that tracking individual processes is going to be an infinite namespace. You would need to have a label for the PID, and the process name, in order to maintain uniqueness. Some systems can generate thousands of processes per second, which is far too much cardinality churn for Prometheus.
If you use a supervisor like systemd or docker, you can use cAdvisor to expose container CPU usage metrics. This would reduce the cardinality to container names, which would be much more possible to track.