As an example on this method, ssh-agent does this. eval `ssh-agent PARAMS`. ssh-agent generates the commands that the shell will then evaluate.
OT: In a previous life I also used this method to set up user accounts at a university to have the proper environment for their classwork (see modules). A user's account, on login, would read configuration files that would generate the appropriate variables, aliases, path settings, and other schtuff specific to their coursework or other projects that they were participating in. The shell (independent of which one they used) would then evaluate these shell commands that were generated. It made my life as an admin much easier, the user accounts much more consistent and predictable, and allowed the professors the ability to request applications and configurations particular (or peculiar) to their classes. Not certain if I would use this particular implementation again, but it did work well in the mid 1990's.