You could perhaps use virtualenv for this, especially if you have a relatively recent version of Python.
That would allow you to create a sandbox environment with whatever packages you need, share it across environments, and run what you need to run.
Another approach (a bit more "complicated"), would be to use a VM solution like Vagrant to set up a sandboxed environment that could be run mostly anywhere. It does require installing and configuration, but it's a one time thing, and can then be used to set up whatever environment you need without messing with the actual system libraries of each machine.
These are just general approaches I would consider, you'll know better what applies to your case since you have knowledge of the actual restrictions and architecture you're using.
I think a virtualenv setup could really be the way to go, you can make it depend on the system packages as a base, and then install tox
on top of the sandbox:
http://www.virtualenv.org/en/latest/#the-system-site-packages-option