Is it okay to use non-parametric tests on normalized data?

I'm doing experiments with 3 conditions: A, B, and C. I have done the experiment 3 times, so each condition gets 3 values for a total of 9 values. Each value is actually an average of 5 measurements. I want to test these data to see if the conditions significantly differ with respect to my response variable. However, I normalize the values from the 3 conditions to the value of A in that experiment. Furthermore, condition C is consistently causing a total reduction of the response variable (an expected result). So for each experiment, the results for A,B,C become 1.0, something between 1 and 0, and 0.0, respectively. I clearly do not achieve the homoscedasticity assumption, so I want to do a Kruskal-Wallis test. Is that okay? Or is my method of normalization violating some Kruskal-Wallis assumption? What if I just had conditions A and C, and wanted to do a Mann-Whitney test?

I'm aware I have low power with such a low N and a non-parametric test, but using my actual data I have a p-value of 0.0667 with only 2 experiments, so I'm fairly confident when I complete the 3rd I will observe significance. • What did you do exactly on "I normalize the values from the 3 conditions to the value of A in that experiment. "? – user158565 Nov 16 '18 at 19:57
• I divided the response variable measured under each of the 3 conditions by the response variable measured under condition A. – user219200 Nov 19 '18 at 3:36