(Originally this answer presumed that by "SCI journals" OP referred to journals published by the Society of Chemical Industry. OP has clarified that they meant journals indexed in the Science Citation Index, I have modified this answer to reflect that.)
Cite articles on scientific merit and relevance alone. You should have a good idea which journals are accepted authorities in your field. You will, naturally, find yourself citing predominantly from these.
Authors, editors and reviewers will often use the reference list as a shortcut to assess whether a particular paper is suitable for submission to/acceptance in a particular journal. They're not necessarily only looking for their own journal, but looking at the fields and pedigrees of the journals cited helps them quickly estimate the field and importance of the manuscript. As an author, you can use this to help you decide which journal you wish to submit a manuscript to; although it will only work if your citation are genuinely made on scientific merit, not on your subconscious desire to submit to Nature! This technique should also help you avoid submitting to predatory journals.
As for citing from dubious journals, this is a thorny issue. One the one hand, one doesn't want to associate oneself with them (or appear to endorse them) and the papers may not be very robust; on the other hand, one does have an ethical obligation to cite relevant prior work - especially where it something you build on or advance past. Various questions on Academia stack exchange deal with this issue, for example:
This paragraph that was included when assuming SCI referred to a publisher. Less relevant to the clarified question, but still relevant to the question of what to cite:
Citing publications from a particular journal or publisher is never a requirement for publication in a legitimate journal. Journals that engage in such practices risk getting themselves lots of bad publicity and removed from indexing systems. Equally, you should have absolutely no hesitation in citing papers from legitimate, non-SCI journals.