This is what I've come up with.
1) Civilian or Law Enforcement unmanned machines - These simply don't possess extreme lethality. These could be designed for surveillance, construction, transport, medical support, repairs, use of non-lethal weapons, etc.
Note: Military versions of these could be upgraded with enhanced protection and abilities such as loading ammunition into a manned machine.
2) Anti-missile/anti-micro-UAV systems, minesweepers, etc. - These only target ammunition too small to be manned.
3) Ammunition / Subsystems - These machines do not operate independently for extended periods of time, and they are activated by a manned system. For those of you who have seen the movie "Battleship", the ball grinder tanks launched by the alien destroyers were an example of highly intelligent ammunition. An example from Starcraft is Spider Mines.
My idea of this third type is an Aircraft Carrier Sending Out an attack wave of UCAVs (Unmanned Combat Aerial Vehicles) on a known enemy location. If they misidentify targets then it's simply collateral damage and not a robot uprising. If their remote command is hacked they won't attack their parent fleet because pre-flight they were given a fire zone around the predicted enemy locations along with a command to return to the carrier within a certain time, and where the carrier might be after that time, and the remote command cannot override this, so hacking will only save the enemy, it will not allow them to retaliate with the UCAVs unless it's a close air support mission, in which case with each wave of UCAVs the carrier would have to take the calculated risk of how likely they are to get hacked.
Note: Optionally-manned machines also fall into this category since you decide when to operate it without a person and could have some sort of restrictions to its operation similarly to as described above.
Have thoughts/comments/questions on the classifications?
Do you think this is all irrelevant because wireless remote control of unmanned combat systems is secure?