I'd go with no.1 because you can find the following minimal pairs in English (RP):
could - cowed /ʊ/ versus /aʊ/
cheese - cheers /i/ versus /ɪə/
dead-dared /e/ versus /eə/
bell-bail /e/ versus /eɪ/
buy - boy /aɪ/ versus /ɔɪ/
etc.
For a comprehensive list of minimal pairs in English (RP) compiled by John Higgins, see http://myweb.tiscali.co.uk/wordscape/wordlist/index.html
Thus, a single diphthong in English represents one phoneme.
This is traditional phonology, not generative.
Naturally, other analyses have been proposed, too. For example, a diphthong is treated as a vowel plus a semivowel or another vowel (McCarthy 1952). Thus, you end up having a diphthong consisting of two phonemes under those approaches. Peter Roach (2009) mentions that this view was "almost universally accepted by American (and some British) writers from the 1940s to the 1960s, and still pervades contemporary American descriptions" (p. 104). Roger Lass (1984) argues that analyzing a diphthong as a vowel plus a semivowel /j/ or /w/ is a "shaky position" (p. 138). He also convincingly dismisses the approach proposed in SPE (Chomsky and Halle), characterizing it as "seem[ing] close to lunacy" (p. 138). Bruce Hayes (2008) concludes that the "fewer-phonemes-the-better" analysis is not an iron-clad argument and that the segment/sequence problem is still "an unsettled one in phonology" (p. 57).