Initial velar and palatal nasals occur in only a minority of the
world's languages. When a language is simplifying its nasal
inventory, these may be eliminated.
For initial velar nasals, merger with glottal stop seems to be
common, e.g. Mandarin Chinese. Sometimes the merger goes the other
way, and initial glottal stops become velar nasals - e.g. some
dialects of Mandarin, and there's an Austronesian example in the
archives. A change to [h] is also reported within the Tai group,
though this is best evidenced for voiceless (or pre-aspirated) velar
nasals.
For palatal nasals, the only elimination route I can recall is merger
with [j]. This has happened in most Tai dialects. The usual result
is [j], but in Lao the merger resulted in palatal nasals. (Lao [j]
derives from Proto-Tai [?j], mostly still written "?y" in Siamese for
words of tone class B.) If Romanian ever had _initial_ palatal
nasals, it too would provide an example of the palatal nasal becoming
[j].
Are there other elimination routes? For example, when final nasals
are being reduced, it seems that both m > n and m > N (velar nasal)
may occur when an opposition m ~ n ~ N is being reduced.