How those data base came in first place? Just take a classic protein as an example and start from the scratch. 

You mean experimentally or in silico? For the first case, the most common approach is purification followed by biophysical characterization. There are some interesting cases on which the exact function of a novel protein was discovered after solving its structure (there's a very interesting article from Shoichet's group on which they used the crystal structure of an uncharacterized protein to find its function by docking all possible reaction intermediates. See http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=pubmed&cmd=Retrieve&dopt=AbstractPlus&list_uids=17603473&query_hl=1&itool=pubmed_docsum )

There is a plethora of bioinformatics methods to extract information from protein sequence. I particularly like those which search for homologs of the sequence of interest and then proceeds by comparing what your protein has to what is important for the homologs. The simplest case is conservation analysis: if you detect that your protein belongs to a given family of enzymes which all strictly conserves some residues which are not present in your protein, that may suggest it is an inactive form or that it does something else with the same fold. Some interesting results can be found if you look for correlated residues. 

Things are a bit more difficult if your protein is so obscure that there is no published information even for their close homologs - typical case is you run an analysis using PFAM and find out it belongs to a DUF (domain of unknown function). However, sometimes such cases can still present something useful, as distant homology detecting methods in some cases can suggest a DUF may be related to another well characterized family.

The first question is - what information do you have about your protein - from what type of organism does it come from, and how does it vary between related organisms.

In the classical cases, researchers most often had an enzymatic activity they could follow through purification to the point where they could obtain at least a partial sequence against which they could raise antibodies.

Nowadays, the more likely case is that you have a putative open reading frame, giving you the option of following expression profiles and tissue distribution through RNA levels (Northern Blots and in-situ hybridisation), and to overexpress your gene in order to produce sufficient amounts for proper characterisation. However, as soon as you have a reliable sequence of your novel protein, probably the first thing to do is a thorough homology search and search for motifs: e.g. the presence of potential signal sequences and transmembrane domains will give you a hint on subcellular localisation, homology of at least parts of your protein to domains of known function. 

Once you have a sequence, you can produce antibodies against synthetic peptide fragments, or you can overexpress your protein in a suitable production host, possibly with affinity tags that help detection and purification.

Thanks Dr. Honegger and Dr. Bleicher for being brief. I have been studying the whole cell proteomic profile of mouse embryonic stem cell by 2-DE. I obtained some of the proteins which are not in the protein databases. So, I  am wondering how can I study about a new protein. By the way thanks again for letting me know the way which I can study.