You can try SLiMSearch or SIRW. Maybe MAST (from the MEME suite) too.

SLiMSearch is available here: http://bioware.ucd.ie/~compass/Server_pages/slimsearch.html Click on the all proteins link from the right and then enter the motif like this 'G...G'.

Here is one I ran earlier - for such a degenerative motif it takes a while: http://bioware.ucd.ie/~compass/cgi-bin/PHP_helper_files/slimsearchInfo.php?jobId=GC-OIO

There is a new version http://bioware.ucd.ie/slimsearch2.html but this is currently limited to human sequences. It does have some other proteomes in there and if you're particularly interested shoot me a mail and I can add more. Human results here: http://bioware.ucd.ie/~compass/cgi-bin/PHP_helper_files/slimsearch2Info.php?jobId=jpytnb

The advantage of the SLiMSearch method is that it will find instances of the G...G motif and rank them assigning significance to each instance based on the relative local conservation.

SIRW appears to be down at the moment but its http://sirw.embl.de.

Not sure about the second part of the question- identifying proline induced kinks.

EMBOSS (http://pro.genomics.purdue.edu/emboss/) has the very useful fuzzpro tool if you don't need to correct for conservation

I tried "Scan prosite" with the pattern G-X-X-X-G but it said:

ERROR: user motif is too degenerated (generates >1000 matches on a sample of 5000 Swiss-Prot r34 shuffled sequences); CANCELLED

You could BLAST query sequence against subset of sequences from ArchDB (see example entry http://sbi.imim.es/cgi-bin/archdb/loops.pl?class=HH-35.1&set=ArchDB95), or equivalent database of loops.