What program are you using? kmeans in R is a pretty decent package for this. When it comes to naming clusters that is usually a choice made by knowing something about the data units in each cluster.

I think that you have to use hierarchical clustering, that what I heard, but honestly I have never used it

Specifying number of clusters is the job of Area specialist, there may be some methods for automating K but the results will be effecting. For labeling you can use Naive Bayes and Decision Tree.

As Asim ullah suggested, you can use Naive bayes and decision tree for labeling your classes, however, number of k depend upon the data you are going to classify. For selecting number of k, you should adopt some random approach as i have heard that its selection process is performed randomly.

Dear Zafar, random approach is not doing well, only the domain expert of that field(data) can suggest number of clusters accurately. i.e. if you want to cluster students data on the basis of marks, so you know how much clusters will be required may be you want to divided it into A,B,C etc grades or some other criteria. 

ASIM ULLAH JAN,

You are right brother but, in different publications i have noticed that k is selected randomly. When number of classes is not known in advance then k should be selected randomly and can be changed later on, however, when classes are known in advance then Random forest or SVM is appropriate approach to adopt.

Regards

Zafar Ali

Thanks Zafar 

Expert heuristic is the best means to determine the number of clusters. There is also a 'Rule of Thumb' that you can use. It states that K is equal to the square root of n/2. 

The following articles may be useful:

http://www.ijarcsms.com/docs/paper/volume1/issue6/V1I6-0015.pdf

http://papers.nips.cc/paper/2526-learning-the-k-in-k-means.pdf

http://www.ee.columbia.edu/~dpwe/papers/PhamDN05-kmeans.pdf

Thank you for your helpful answers.