The fact that Chern classes are Hodge classes (and are rational combinations of algebraic cycles) is a part of the proof of the "Gauss Bonnet theorem" (as given in Griffiths and Harris). So my question is why is the fundamental class of every algebraic variety a rational combination of them?
I am a novice, but here is what I gather from perusing the literature. The Hodge conjecture asserts that all Hodge classes are spanned by algebraic classes. The fact that all algebraic classes are spanned by the chern classes of (holomorphic or algebraic) vector bundles, is proved by resolving the structure sheaf of a subvariety by a finite complex of such vector bundles. I.e. one defines formal "K groups" generated either by isomorphism classes of locally free algebraic sheaves (vector bundles), or more generally by coherent algebraic sheaves, with an equivalence relation defined by formal alternating sums of sheaves occurring in exact sequences. The fundamental result that all coherent algebraic sheaves have resolutions by locally free algebraic sheaves shows these two K groups are isomorphic. Hence the subgroup of algebraic cohomology classes, which is the image of the chern character map on the K group of algebraic coherent sheaves, equals the image of this map on locally free sheaves as well. Since the chern character is generated by chern classes, the result follows. The 1974 book Topics in algebraic and analytic geometry, Princeton mathematical notes #13, by Phillip Griffiths and John Adams discusses this in detail.