Thank you for your answer. I asked this question because I collected such a sever skewed data for my PhD. I used PLS first without any normalization for it as the literature said that PLS is a distributional free statistical method. The results was poor and I got a little significant relationships which was not sufficient to test my research hypotheses. Then I asked help from a professor in statistic and he advised me to normalize my data set first then run the data. I argued with him about the ability of the PLS to run non normal data as literature said and he tell me that some data sets can be described as noisy or sever skewed so it is in need for normalization. I used some algorithms to normalize all sever non normalize items within my data set. Next, I run the data again and surprisingly I got a very good results !! Now I really think that PLS is indeed not a distributional free method and asked researchers to reconsider this assumption. Am I right? I am not sure!!!