I have a gene expression data including intrinsic subtypes of breast cancer, but it does not have paired normal samples, Is it possible to use normal samples from different batch if I use batch effect removal methods eg COMBAT or DWD?
No, that's not possible.
Your batch effect removal method doesn't know what is "technical-batch" difference and biological difference.
So when comparing tumor samples from batch A with normal samples from batch B you might be comparing technical differences in batch A with technical difference in batch B. Not biological differences. You don't know how big the batch effect is.
If you would have 4 tumor and 4 normal samples in batch A and 3 tumor and 3 normal in batch B you could combine the data and attempt batch removal, because now your software 'knows' which differences are biological and which are technical.