Yes, it certainly can. Small corpora are often used to carry out research into specific genres. However, there are a couple of points that are worth bearing in mind,. Firstly, is the corpus big enough to generate sufficient examples of the structures that you intend to analyse? Rarer strucutures will require larger corpora, as will structures that enter into a wide variety of phraseological patterns, such as some and any. Secondly, you have to look very carefully at the composition of the corpus in order to be certain that the corpus you choose is truly representative of the genre you aim to investigate.
The Corpus Linguistics in the South Conference held at Sussex University in February of last year was devoted to the whole subject of small and large corpora. It might be worth your while to have a look at the talks from the conference that were on small corpora specifically.
If a corpus is too small than you can call it regular datum. Typical corpus is expected to have big data.
A specialised corpus doesn't have to be too large (e.g. the type I compiled - "corpus of Nigerian media language," which perfectly served the purpose). I recommend a similar enterprise if necessary.
The size of the corpus is related to the phenomena or phenomenon you want to illustrate, as Mario Bisada pointed. This is not an easy problem at all. Can you develop your objectives some more? What do you mean by "structural linguistics patterns" ? And also " genre"?
It seems to me you want to explore something related to collective style, and the balance of your corpus thus becomes crucial. Do you want to contrast one or more genres? That would be easier on small corpora provided they are far apart. Do you have a reasonable hypothesis on the linguistic patterns that would characterize one genre or another? From my experiments, genre needs fairly big corpora. Personal style will interfere in your data in a small corpora, blurring the genre archetype.
But then it also depends on the tools you will rely on. The sharp contrast between Chris Turner's and Hendi Pratama's answers can be explained by the differences in methods: data mining requires big data, while search for descriptors can hardly be handled with them. Smaller corpora are needed if you have to manually sift and screen results in different corpora, later trying to test on bigger corpora.
There are still few significant methodologically sound research results that would help.
It is worth spending time to explain your experiments and choices. Typically, lack of rigour in the corpus choice including its size do jeopardise results in too many cases, because the experiments cannot be reproduced nor checked and therfore are nor cited.
A few references on that point will follow if you are interested.
Thank you very much, Nadine. Your comments are very useful.
Micro-level attempt will yield result depending upon your analytic capacity