@Anjo I think the point is that depending on the settings Conda Environment Preopagation would try to reproduce the exact packages and versions of the Python modules wich tend to differ depending on the operation systems. I am not sure about if it only checks the names (in theory that should cover it).
What I would do is select a good basic conda configuration for KNIME (YAML file) suited for your operations system and basic thing you want to do (namely KNIME based Deep Learning or not which requires specific settings). You will have to settle on a major Python (3.6, 3.7, 3.8 …) version (the best one compatible with your specific package - in this case RDKit).
Then you should maybe limit the channels to “conda-forge” and maybe some additional pip installations.
If your special package would not immediately install via YAML that is fine just add it later. Once you have done that on your specific operationg system you can then store the whole thing in a Conda Environment Propagation and deply that to other people.
You might get an idea from this example:
We all look forward to the deeper integration of Anaconda/Python and KNIME that has been announced.