I am tempted to use SVA for removing batch effects from two different technologies from which I get counts data for miRNA. The data do not have common reference (different treatment in both technologies).  I understand concept of batch effect (days, reagents, technicians etc ...)  BUT am I allowed to use SVA if 2 different technologies were used to get these count data?

Probably not. In addition, if the experiment isn't completely duplicated on the two platforms and you are intending to make comparisons between the two platforms, then you are highly likely to get lots of differences that are not biological, but instead are technical.

If the experiment is duplicated on the two platforms (say you have a 3 x 3 treatment/control experiment using a HiSeq, and another 3 x 3 using an Ion Proton), then you could consider using something like the GeneMeta package.
 

You could probably use SVA to correct for the unmodeled confounders (e.g. the different technologies). However, James makes an important point - you need to (a) make sure you have figured out how to normalize the data in the two platforms to make them comparable and (b) you need to make sure that technology and biology are not completely confounded. For example, if you measured all of one type of treatment with one technology and all of the other kind of treatment with another technology, then you have what is called a perfectly confounded design. In this case it is *very hard* to distinguish biology and technology effects. This blog post does a good job explaining the issue: http://simplystatistics.org/2015/05/20/is-it-species-or-is-it-batch-they-are-confounded-so-we-cant-know/

Thank you for your insides. Right, I think we have just enter to the era of PCD (perfectly confounded designs) , even here it was obviously a badly planned experiment that I was given to analyse.