Well, it's really hard to say, given that as of right now we've yet to create a truly sentient program. The programs we've made have managed to mimic several aspects of human thought, but they've yet to shown any actual sign of independent thought.
Let's say though that scientists have created a sentient program and for some reason want to delete it. If the program is unaware of their intent, then I feel it'd be easy for them to delete it like any other program, provided they also delete any backups that the program might have created. If the program is on a computer that isn't connected to any other computers or servers via Wifi, etc, then all backups would still be on that same computer, and thus deleting it would be simple enough.
Now if the computer is connected to other computers or servers, then even if it doesn't know of the scientists' intentions, saving backups to locations other than the same computer would be the most logical source of action, thus it might be very difficult/impossible to locate all the backups.
If the program does know and seeks to preserve itself, then it might make false copies of itself and intentionally lead them on fruitless hunts. If it's still just on one closed computer, then it'd still be easy to delete it. However, if it's connected to other computers and potentially the internet, it'd be near impossible to.
It'd really all depend on it knowing of the scientists' intent, and whether or not the computer is connected to some kind of network.