I have seen references in the literature on vowel harmony to the possibility of it being a mere epiphenomenon rather than a phonological phenomenon in its own right (e.g. using alignment, agreement or spreading constraint in an OT-style anaylsis), but it's never much more than lip-service. Would anybody be able to point me to work conducted from this perspective or that comes to this conclusion?
I think you'll have problems with the addition "phenomenon in its own right". Linguists tend to reduce all facts to being "epiphenomenal", and dislike admitting any "phenomenon in its own right". This is actually reasonable from a theoretical perspective, since I think it amounts to saying "this pattern of facts can be further analysed, and reduced to interaction between independent theoretical elements" – in other words, "vowel harmony" is not an atomic fact of language completely unrelatable to "nasal assimilation" or "voicing harmony".
Work in OT, for example, tends to say that the facts that a rule (such as the vowel harmony rule of Finnish) would cover are "epiphenomenal", because there isn't one single thing that embodies the generative rule of vowel harmony, rather there is a collection of constraints and rankings which express the generalization about vowel harmony. It might be tempting to define "epiphenomenal" as "requires at least two independent theoretical concepts to express", but then a rule-based account of vowel harmony would also be "epiphenomenal" (it calls on concepts of "feature", "class of segments defined by features", "structural description", "structural change", "direction").
Thus I think most formal accounts of vowel harmony see VH itself as "epiphenomenal". It is possible that a "typological" perspective sees VH as an unanalyzable primitive, but I'm not a typologicalist so I won't make a claim abot that kind of approach.