Languages like Japanese are said to have case phrases (KP). I don't understand why case particles are considered heads. Why is the structure of NP+ga [KP [DP x] [K ga]] (ie the DP depends on ga)?
I don't have detailed knowledge of Japanese but in Urdu there are two approaches. One says that the case markers are suffixes, whilst the other assumes that they're particles and posits a KP. A KP in Urdu is something like a PP but it's headed by a case marker (K). In fact, there's a hierarchy: PP>KP>DP>NP.
Note that the categorial head of a phrase needn't be its functional (lexical) head. Functional heads are content words.
Japanese doesn't have DPs. Those words that can appear in the position associated with determiners/articles, are clearly marked for attribution. The demonstratives kono, sono, etc. contain the genitive particle -no. Others such as iwayuru 'so-called' contain attributive suffixes, here: -uru.
Concerning your question, the NP depends on the case particle, because the presence of the case particle is the most important criterion for the grammatical distribution of the entire expression NP-Case. Look at the next example:
(1) Kare-wa Tokyo-e modotta. he-top T-all return.pst 'He returned to Tokyo.' (2)*Kare=wa Tokyo-o modotta. -acc
In (1), the directional complement Tokyo is properly marked by the allative case particle -e (the dative particle -ni is also possible). Hence the sentence is good. Example (2) shows the accusative case particle -o instead of -e. The result is bad.
If the case particles weren't somehow head over the NPs, that effect would be difficult to explain. In fact, assuming the following structure for Tokyo-e:
allows one to generalize over the noun Tokyo, and to arrive at the more abstract next structure:
That means that any noun that references a place could appear instead of Tokyo in (1).
Assuming that the place noun is head over the case particle would lead one to:
Structure (5) views case particles as subordinate to their nouns. As a result, any case particle should be good. But example (2) shows that this is not the case.
In summary, sentence (1) isn't good because Tokyo appears; any place name or local noun would be fine. Rather the sentence is good because the allative appears, rather than another case particle (apart from the dative).