I don't know what to call this yet.
One definitive aspect of self-replicating machines is that they are able to self-replicate, meaning that all of its function is self-referential in that the dependencies are 'internal' -- the design of the system could be envisioned as having "dependency loops" between the functionality. There is no functionality that is of the self-replicating machine that is not self-replicable itself. In other words, if when the machine replicates and makes its duplicate through its own internal functionality, some function is dropped then that 'some function' was not replicably definitive of the system.
An interesting exception (2008-06-23 -- everything above is from earlier thoughts in March or April of 2008); if the overall replicable system is able to produce something that can make a functional replica of another function that would otherwise not be immediately replicated, then does this function get included in the definition / elaboration of the original system in the first place? In other words, is can a suitable substitute for will? In the sense that the system will completely self-replicate versus well, sure it can?
Another issue is that designing self-replicable processes/systems is that it is like poking a very specific target through a very very tiny pinhole on the size of a few angstroms, and you have this giant baseball bat and you're swinging away at it. Given some object that you want to replicate, like a rock, you can't just ask it nicely please, would you be so kind as to proceed to duplicate yourself? This isn't going to work out. You can try. I am not going to stop you. But it will not work. Maybe that's a good television show to try out? The Rock Whisperer. I don't know what it would be about. It's just a rock sitting there. The functionality (indeed what little a rock has in the first place) isn't going to allow it to self-replicate, and the basis/design of that object's (rock's) functionality isn't necessarily ever going to be extendable to an extent to allow self-replication.
So, if you give the system some parts anyway, parts that it doesn't automatically make, and as long as that system can replace those same parts later in its life cycle, then this is kind of like "replicable-integration", yes? Does it then qualify as self-replicable?