Current archaeological theories on religion in Gaul and the other northwestern provinces emphasise the creative hybridisation of belief and practice; in other words, the creation of new gods and forms of worship by fusing together elements of both pre-existing Gaulish and incoming Roman religion. This is part of wider movements in archaeology and ancient history that seek to do away with older ideas of Romanization, which suggested that cultural change in these provinces was monodirectional, irresistible and planned by the Roman authorities themselves.

The most obvious change is in the way that deities were depicted. Pre-Roman images of gods were very abstract, emphasising specific elements like heads or horns, rather than creating realistic portraits. Under the Romans, Graeco-Roman art styles were applied even to traditional Gaulish gods, including Epona whom you mentioned in your OP. It's also only under Roman rule that inscriptions mentioning the names of these deities appear - there was no culture of literacy in Gaul before the Roman conquests. This is one of the factors that makes it very difficult to reconstruct pre-Roman religion, and to work out which gods were already being worshipped and which only became popular later.

There is also the issue of interpretatio Romana, a process by which native gods were associated with Roman ones. As both the Gaulish and the Roman religious systems were polytheistic, they could easily make room for new deities. An example of this pattern is Lenus Mars (http://www.bbc.co.uk/ahistoryoftheworld/objects/oQ5HoARDSOiWFTJlqwChNQ). We also find 'divine marriages', where a local goddess is paired up with a male Roman god. Mercury and Rosmerta are the most common pairing, but there are others.

Exactly where the impetus came from for these associations and marriages is a matter of debate. Older scholarship saw it as evidence of Roman culture being actively spread to the natives, and there's quite possibly an element of that, especially in large towns with a more concentrated Roman presence. It could be a display of dominance over the conquered peoples, and a way of integrating them more closely into the Roman imperial system. More recently, archaeologists and historians have been asking why native people would go along with this kind of thing. To a certain extent, and particularly for the local elite, staying close to the Romans was a way of maintaining their position and accessing all the new power and wealth now available. Participating in Roman forms of worship would've helped in that effort. However, there must have been other motivations for people further down the social scale. For them it was perhaps a matter of trying to work out their place in the new Roman world, negotiating between existing Gaulish customs and those imposed by their Roman overlords. Webster recently suggested that the divine marriages between Roman and Gaulish deities might even be a covert sign of resistance to Roman rule. Although the Romans may have interpreted these pairings as male dominance over a provincial female, the Gaulish system seems to have invested more power in the feminine, therefore interpreting a marriage between Rosmerta and Mercury as the latter being subservient to the former.

In terms of actual forms of worship, again we seem to see the the blending of Roman and Gaulish practice. Before the Roman conquests, it's possible that headhunting and human sacrifice were prominent ways of honouring the gods, and although the Romans but a stop to such practices, archaeologists have noted that the emphasis on heads seems to remain. In places like Chamalieres, model heads were still being given as gifts to the gods well into the Roman period. Pre-Roman traditions also included the giving of weapons as offerings, perhaps connected to the privileging of warfare in Gaulish society. Under the Romans, the offering of full-sized weapons gave way to miniaturised versions, which has been interpreted as an adaptation of the traditional practices to new social contexts, in which inter-tribal warfare was no longer a feature.