Using GFP fusions can be tricky. Have you thought about using a smaller fusion such as V5 or FLAG tag vectors. This may eliminate the potential effects of GFP on protein folding or function. Alternatively, GFP could be added to the C-terminus of the protein.

Tagging with GFP can certainly disrupt the function of the protein you are studying. It could be that the protein is non-functional, or a dominant-negative, or have altered function, or be perfectly fine. It may also affect the interaction of the protein with other binding partners within the cell, or trafficking. Assessing the function of the GFP-tagged protein in some sort of in vitro or heterologous system is advisable. Immunocytochemistry could be used to check if localization is the same as native protein. "Dots" could be aggregated protein/protein that has entered a degradation pathway via proteasomes or lysosomes. Checking for a ubiquitination "smear" on a Western might be helpful. Also be careful if your protein has an N-terminal signal sequence which would cleave the GFP. Does a C-terminally GFP-tagged protein look the same?

All tags have the potential to disrupt or inhibit native protein functions. If you have an ER resident protein, it is important to know if it has 1) an N-terminal signal peptide and a stop transfer that acts as membrane spanning domain (type I membrane protein), 2) just a transmembrane domain without a signal peptide (can be either type I or type II dependent on the charge distribution of amino acids flanking the transmembrane domain), or if it is a multiple membrane spanning protein... If it is type 1, you might want to consider a C-terminal fusion (which leaves the GFP in the cytosol), but you have to make sure the ER retention motif (KKXX) is still C-terminally exposed. If it is type II, you can have the GFP in the lumenal domain (this is where the C-terminus will be) without affecting ER retention motifs (which are exposed at the N-terminus). If it is multiple membrane spanning, you may have to try N-terminal, C-terminal or inside a cytosolic loop. If it is a soluble protein with a signal peptide and a KDEL/HDEL signal, then you can try to fuse the GFP between the signal peptide and the N-terminal portion of the mature peptide, or between the KDEL/HDEL and the rest of the protein at its C-terminus. Some GFPs and YFPs are known to form dimers, and that could stop your attached protein from working, so using a monomeric fluorescent protein is also important. A lot of things can go wrong and unless you have the crystal structure of your protein, you cannot make any prediction whatsoever. So you can only try and see.

Thanks everybody for your help. I'm glad to update you on my work.

I couldn't use flag or any tag because I have to be sure the cell I'm analyzing in vivo (so no possibility to do immunoistochemistry) is expressing my protein, that's why I used gpf fusion. 

Dealing with C-terminal fusion, I have tried to inject a plasmid encoding for PS2 fused to GFP to C-terminal, but no fluorescence was observed.

The protein I'm studying is a ER protein with many membrane spanning domain (9). I successfully performed a immunoistochemistry against the protein (the same one encoded by the UAS-GFP-protein, but without GFP) and I confirm dots along the axon. So I believe these dots are specific for my protein and not due to GFP toxicity. 

Since the neurons I'm studying are Rohon-Beard sensory neurons, does anybody knows if these neurons are myelinated one?