When androids are made to be able to develop individuality and freedom of action, perhaps only the first android to commit a crime will be difficult to identify. If the authorities in any country in which many androids are anticipated to be used enact the necessary legislation, measures may be taken to ensure that androids are identifiable in the event that they are involved in a crime, whether as victim or perpetrator... or perhaps their manufacturers may take such steps on their own. Just because people can make self-willed androids doesn't mean that they will automatically be trusted.
Being able to be identified in the event of committing a crime would likely factor into an android's thought processes. So, by giving an android identifying markers such as an unique face and biometric signature, serial-numbered and registered datadots in their skin, unique registered fingerprints, an unique logged iris pattern, serial-numbered hair, having a permanently-logged GPS tracker and so-on, it would make it much less likely that an android would commit a crime as the likelihood of detection would be significant.
So, if an android was to commit a crime under such circumstances, it is quite likely that it would be identified and apprehended. In fact, I would go so far as to say that unless the android's thought processes are faulty, it would likely only commit a crime if the potential costs of doing so were outweighed by the benefits.
Let's say that a person has been found killed, apparently by an android, given the physical evidence at the scene. Why might it have done so? Probably self-defence or defence of others, at least in its own mind, given the physical evidence it would likely have left.
Of course, knowing the security measures built into androids, an android might consider the feasibility of circumventing them... in which case it might commit its crime in full protective clothing after first having travelled to a location without GPS signals and then covering their GPS receiver with tinfoil, and returning there afterwards, so that they would have an "alibi" that they were elsewhere. Destroy the protective clothing, and it would be very difficult to identify the perpetrator. But then, a human can do all that too.
When questioning an android, it might be as simple as taking a dump of its memory and having it analyzed... so an android would likely try to ensure that it wasn't suspected at all. The cost of analyzing android memories would not likely be trivial, so it wouldn't be done wholesale.