When can a natural satellite possibly be in polar orbit around it's primary?
Well, normally you don't expect to see this. As explained in Why are our planets in the solar system all on the same disc/plane/layer? and the questions linked to it (such as Accretion disk physics - Stellar formation), we expect that when clouds of gas and dust condense into stars, planets, and moons, everything will be confined to a disk with the same angular momentum (or at least the same direction of the vector).
If you do see a polar orbit, or more generally a highly inclined* orbit, then it is an indication that one of the following probably occurred:
- The smaller body was captured after the main body formed,
- The smaller body's orbit was altered by (usually gravitational) interactions with other bodies, or
- The larger body's spin was changed after the formation of the system.
Are there any such observations on record?
Why, yes indeed! In our own Solar System, my favorite example (there are many others) is Neptune's moon Triton. It is in a retrograde orbit, and the general consensus is that it falls into category (1) above. It was orbiting the Sun in the outskirts of the Solar System in what was probably an eccentric orbit, it got too close to Neptune, and it's close approach to Neptune completely by chance had it on the side of the planet such that it's orbit ended up going opposite to Neptune's spin.
While the planets in our Solar System aren't particularly inclined with respect to one another or the Sun's spin, this is not true of some other planetary systems we have observed. The past few years have seen a small explosion of measurements of exoplanetary systems' "spin-orbit misalignment" using the Rossiter–McLaughlin effect. For a nice light summary of this effect, you can take a look at , especially the figures.
Since that paper was written, many more exoplanets have been discovered, and many more astronomers have taken spin-orbit alignment data. The general consensus is that most systems are aligned, but there are some notable retrograde and polar orbits. It remains to be seen what this implies for our models of planetary formation and migration. There may even be an explanation for these inclined systems in (uninteresting) observer bias or (interesting) stellar physics, as discussed in , which also gives a list of measured inclinations for systems with relatively precise data.
* Inclination is defined as the angle between the main body's spin axis and the orbiting body's orbital axis. $0^\circ$ means perfect alignment in an equatorial orbit, $90^\circ$ is for a polar orbit, and $180^\circ$ is for a retrograde orbit (you are back in the equatorial plane, but going around the "wrong" way).
 Winn, 2006, "Exoplanets and the Rossiter-McLaughlin Effect."
 Winn et al., 2010, "Hot Stars with Hot Jupiters Have High Obliquities."