Tasks 1-4 scream XML::SAX to me (which is kind of like dragonchilds suggestion of "XML::Parser in stream mode", but different). I say this because none of these tasks require you to load the entire XML document into a tree, nor do they require a heafty amount of contextual knowledge either.

The nice thing about XML::SAX is that it can be PurePerl, or it can use a C based parser if it finds it. XML::Simple, while a very useful module, can be slow, and will build a structure of the entire XML doc for you, which is not what you need.

Task 5 requires XPath, and I know very little about XPath, and even less about what modules support it, so I am afraid I cannot help there.